[c-nsp] Best practice for redundancy

[Phil Mayers]

Asbjorn Hojmark - Lists wrote:
>> If anyone has experience of UDLD succeeding or failing to 
>> detect errors on links, I'd like to hear about them.
> 
> In my experience it works fine.
> 
> To test it, take two active (up/up) links and swap the two rx
> or tx fibers at one end. (UDLD also works for defektive GBICs,
> but you may have trouble finding one).

Oh, it certainly works very well for that class of "hard" errors, and is 
on by default on all our links that will support it.

I was thinking more pernicious errors e.g. high bit-error rates where 
the link will pass traffic, but much of it ends up discarded, which is 
particularly disastrous with the various load-balancing schemes used for 
aggregates - approx. 50% of your hosts at the other end of the link end 
up with symptoms similar to duplex mismatches! We've seen fibre patches 
(particularly single mode) occasionally go bad and exhibit these symptoms.

Running multiple layers of protocols e.g. UDLD, BFD, OSPF, BGP is 
clearly more likely to fail with such a link than one layer, so arguably 
running as two separate layer3 links might detect such faults faster. 
Frankly I doubt it's worth the marginal gain, but was wondering if 
anyone had such comparative experience.